Research On The Watershed Algorithm Based Object Segmentation

Recent years, with the widespread popularity of computers and the rapid development of modern information technology, people increasingly resort to the computer to a large number of rapid acquisition and processing of visual information. People are, in the process of processing and researching of visual information, often interested in some of thespecial areas in the image or video, and these regions often contain specific and unique property of the visual information. Before these regions can be further handledand analyzed,they should be clearly and precisely separated into single regions. The technology of target region extraction is exactly the visual information segmentation technique. As the basis of processing visual image information,the segmentation algorithm applied in image and video is the key technology in visual information processing.Mathematical morphology is a new theory of nonlinear image(signal) processing and analysis, it does away with the traditional view of the numerical modeling and analysis, characterizing and analyzing images from the collection point, and provides a unified processing tools for visual information processing problems because of its complete theoretical system and intuitive geometry performance. In these thesis, we combined mathematical morphology theory with classical watershed segmentation algorithm and applies in image and video segmentation. Currently,watershed algorithm mainly focuses on two aspects:the difficulty in effectively determine of initial seed, and the problem ofover-segmentation. This paper focuses on the two respects to conduct in-depth researches, by methods of mathematical morphology and other tools, improving regional watershed transform segmentation algorithm. Works and researches of this paper are as follows:1. Elaborates research background and its significance, provides the concept of the video image segmentation, summarizes the properties of the analysis and segmentation algorithm and research status. The basic theory and application of the algorithms are elaborated in this article.2. Traditional watershed algorithm is vulnerable to adverse factors such as light intensity, noise and other interference. With some image pre-processing work, including image enhancement, median filtering, edge detection and mathematical morphology opening and closing (closed to open) reconstruction filter and other operations, we effectively achieve the purpose of noise suppression, and laid a good foundation for the follow-up work.3. In the application of image segmentation, traditional watershed algorithm faces the problem of how to effectively select the catchment basin.Through the early pre-processing and adaptive regional mark extraction algorithm, we alleviate the difficulty that catchment basin can not be automatically identified and generated, and can quickly and accurately calculate the regional mark as well as overcome the impact of noise in the images; the tool of Mumford-Shah energy equation is also used. It achieves the merging of the over-segmentation areas for traditional watershed algorithm on the problem of over-segmentation. The improved energy function lowers computational complexity, achieves simple parameter selection and greatly accelerates the combined speed of over-segmentation area. Comparison of experiment with compared algorithms shows that the proposed algorithm can acquire more accurate segmentation results.4. experimental summary. Some of the major comparison and analysis of the experimental results are inclued in this part. Apart from that, the focus and difficulty of work on this subject in the next step is described, and the prospect of the development of the watershed algorithm based on mathematical morphology is also depicted.